[mattwthompson]

Closes #208
Closes #213
Closes #226

Part of #148

[codecov-commenter]

Codecov Report

❌ Patch coverage is 48.21429% with 29 lines in your changes missing coverage. Please review.
✅ Project coverage is 82.15%. Comparing base (ae11c3b) to head (d64c087).

Files with missing lines	Patch %	Lines
tyff/compute/_openmm.py	39.13%	14 Missing ⚠️
tyff/compute/_produce.py	14.28%	6 Missing ⚠️
tyff/compute/_equilibrate.py	20.00%	4 Missing ⚠️
tyff/compute/_minimize.py	66.66%	2 Missing ⚠️
tyff/compute/workflow.py	66.66%	2 Missing ⚠️
tyff/targets/new_thermo.py	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #209      +/-   ##
==========================================
+ Coverage   82.11%   82.15%   +0.04%     
==========================================
  Files          58       59       +1     
  Lines        5311     5323      +12     
==========================================
+ Hits         4361     4373      +12     
  Misses        950      950              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[mattwthompson]

With the code at 8de767f I can run three epocs with the loss function slowly decreasing (as calculated) considering both thermo and energy targets (not entirely validated)

#213 needs to be resolved before considering this complete, though, and I'll need to re-run

[mattwthompson]

After applying several fixes (using the correct reference force field, getting each loss component roughly of similar magnitude, pressure handling #215, respecting SMILES mapping #219 (comment), and likely more I'm forgetting) I'm still seeing the loss grow in the first few iterations:

EPOCH 0: loss=19.721139
	loss component energy (after applying COEFFICIENTS={'energy': 0.1, 'force': 0.01, 'thermo': 100000.0}
	energy 4.219335556030273
	force 2.288978872135019
	thermo 13.212824719124328
EPOCH 1: loss=23.258530
	loss component energy (after applying COEFFICIENTS={'energy': 0.1, 'force': 0.01, 'thermo': 100000.0}
	energy 4.227046966552734
	force 2.2833941983519064
	thermo 16.748089025622193
EPOCH 2: loss=25.659948
	loss component energy (after applying COEFFICIENTS={'energy': 0.1, 'force': 0.01, 'thermo': 100000.0}
	energy 4.232955455780029
	force 2.280235880638339
	thermo 19.146756829684364

This could easily be because of

• dhvap/dhmix not being incorporated
• re-running all simulations each epoch with no shortcuts/re-sampling/etc.
• equilibration and sampling times being very very small
• uncertainty not being accounted for

or simply expected behavior given the very small number of iterations

---

Update: Running it a bit more, it seems like the behavior is neither ideal nor optimally bad, which in this scope feels good enough:

EPOCH 0: loss=19.721139
	loss component energy (after applying COEFFICIENTS={'energy': 0.1, 'force': 0.01, 'thermo': 100000.0})
	energy 4.219335556030273
	force 2.288978872135019
	thermo 13.212824719124328
EPOCH 1: loss=23.277236
	loss component energy (after applying COEFFICIENTS={'energy': 0.1, 'force': 0.01, 'thermo': 100000.0})
	energy 4.234147548675537
	force 2.2949994226426402
	thermo 16.748089025622193
EPOCH 2: loss=25.690416
	loss component energy (after applying COEFFICIENTS={'energy': 0.1, 'force': 0.01, 'thermo': 100000.0})
	energy 4.2439494132995605
	force 2.299709611384476
	thermo 19.146756829684364
EPOCH 3: loss=19.928845
	loss component energy (after applying COEFFICIENTS={'energy': 0.1, 'force': 0.01, 'thermo': 100000.0})
	energy 4.252472400665283
	force 2.3041207905795678
	thermo 13.372252127332729
EPOCH 4: loss=23.636336
	loss component energy (after applying COEFFICIENTS={'energy': 0.1, 'force': 0.01, 'thermo': 100000.0})
	energy 4.25861120223999
	force 2.3058810113260924
	thermo 17.071844071399056
EPOCH 5: loss=22.374276
	loss component energy (after applying COEFFICIENTS={'energy': 0.1, 'force': 0.01, 'thermo': 100000.0})
	energy 4.262715816497803
	force 2.3058808461727764
	thermo 15.805679609410818
EPOCH 6: loss=31.239997
	loss component energy (after applying COEFFICIENTS={'energy': 0.1, 'force': 0.01, 'thermo': 100000.0})
	energy 4.266695499420166
	force 2.306119460100159
	thermo 24.667181566169205
EPOCH 7: loss=16.541983
	loss component energy (after applying COEFFICIENTS={'energy': 0.1, 'force': 0.01, 'thermo': 100000.0})
	energy 4.2703070640563965
	force 2.3067892802177674
	thermo 9.964886455025024
EPOCH 8: loss=12.080359
	loss component energy (after applying COEFFICIENTS={'energy': 0.1, 'force': 0.01, 'thermo': 100000.0})
	energy 4.273726940155029
	force 2.307612904638319
	thermo 5.499019234473934
EPOCH 9: loss=12.896108
	loss component energy (after applying COEFFICIENTS={'energy': 0.1, 'force': 0.01, 'thermo': 100000.0})
	energy 4.275704860687256
	force 2.3072598936003312
	thermo 6.31314277199202

[lilyminium]

A couple minor comments but otherwise LGTM! I could reproduce locally. I did run into some ManagerLostErrors but they seemed sporadic and went away when I reran.